Trust is an inequality, not a virtue

Whether your distributor keeps your money next month is not a question about your distributor. It is a comparison between one number and another: what taking the month is worth to them today, against what the rest of the relationship is worth to them from here. Trust is the output of that comparison, not an input to it.

Which means the lever is never the conversation. The lever is one of four variables, and this piece tells you which one to pull.

Why these three models

The event is ordinary: you are deciding how much exposure to give a counterparty who could take it. The features that fire are a repeated relationship with a next period, a signal you do not control, and a decision that has been made many times before by many other founders.

Three models, three different kinds of error. The patience test tells you whether cooperation can hold at all, and it produces an equilibrium answer. The monitoring layer tells you what your counterparty can actually see, and it produces a complex answer where punishment lands on people who did nothing wrong. The reference class tells you how often this arrangement fails for everyone else, and it produces a cycle answer that ignores the story you are currently telling yourself. Those three error structures are close to independent, which is the only reason running all three beats running the best one.

1. The patience test: can this hold at all

The formal object underneath every relational contract is an inequality. Cooperation is sustainable when the discounted value of continuing exceeds the one-shot gain from defecting. Write it as future stream against immediate grab and you have the whole apparatus.¹

Four inputs. The per-period value of the relationship. The size of the grab available in any one period. The counterparty’s discount factor. And the outside option that sets the floor under all of it.

Only one of those moves much, and it is the one founders misread. The discount factor is not patience as a temperament. It is the horizon the counterparty perceives. This is the point Osborne and Rubinstein make and Mailath and Samuelson quote at the head of their chapter on finite horizons: a model should capture the horizon the players perceive, not the one that physically exists.¹ If the end approaches slowly enough to be ignored, people behave as though the relationship runs forever. When the end becomes visible, finite-horizon logic switches on and cooperation unravels backwards from the last period.

That is why a five-year supplier defects in year five and not in year one. Nothing about their character changed. Their horizon became visible.

The patience test, the sustainability lens

• Assumes: both sides compare a stream of future value against a one-time gain.
• Fits because: the relationship has a next period and the grab is available now.
• Breaks when: monitoring is private, so the two sides see different things and cannot coordinate on what happened.
• Evidence: grade A as a diagnostic. Barred as a forecast: the folk theorem shows almost any outcome can be sustained, so it predicts sustainability and never selection.
• Counteracts: the belief that relationship quality is about rapport.
• May reinforce: a cold reading that misses genuine reciprocity where it exists.

2. The signal layer: what they can actually see

The second model is the one most founders have never been given, and it explains the injuries that feel most unfair.

In the clean version of a repeated game, each side observes what the other did. In the real version, each side observes a noisy signal of what the other did. A late delivery caused by a port closure and a late delivery caused by your negligence arrive at the buyer as the same thing.

The consequence is not that the buyer is being unreasonable. The consequence is structural, and Mailath and Samuelson state it flatly in a section they title “Punishments happen”: under imperfect public monitoring, the punishment phase is entered repeatedly over the course of equilibrium play, and it is entered in periods when both sides know that nobody deviated.¹ The innocent party complies with the punishment because, given that the other side is entering it, entering it is the best available response.

So the founder question is not how to prove innocence. Innocence is not what triggered the penalty. The question is what second signal you can put in front of them, because the same source shows that increasing the precision of monitoring expands the set of outcomes that can be sustained at all.¹ Observability is not compliance theatre. It buys you deals that are otherwise unreachable.

The signal layer, the observation lens

• Assumes: your counterparty sees a noisy public signal of your conduct, not your conduct.
• Fits because: you carry outcome risk you do not control.
• Breaks when: monitoring is private rather than public, in which case coordination fails in ways this model does not capture.
• Evidence: grade A. Structural, and confirmed in field settings where deliveries are observable and effort is not.
• Counteracts: the instinct to argue that you were not at fault.
• May reinforce: fatalism about outcomes you could in fact have controlled.

3. The reference class: what usually happens to arrangements like this

The third model refuses your particulars. Before you estimate whether this counterparty will hold, ask how often arrangements with this structure hold for anyone.

The empirical literature on relational contracting is unusually well matched to African and other thin-enforcement markets, because it concentrates on exactly those settings: informal arrangements sustained by the value of future interactions, in places where formal contract enforcement is weak.² Two findings from that literature belong in every founder’s base rate.

The first is that relationship value is a function of relationship age. In the Kenyan rose export sector, where flowers are perishable and both sides are exposed to opportunism, the value of a relationship increases with its age, and reliability measured at the moment of a shock predicts both future survival and future relationship value.³ Age is not sentiment. It is accumulated evidence about type.

The second is sharper and cuts against conventional advice. In the Rwandan coffee chain, bilateral trust correlates with relational contracts and generalised trust does not.² Trust in general is not the asset. Trust in this specific counterparty, backed by this specific future, is the asset. Money spent building a reputation for being trustworthy in the abstract is money spent on the variable that does not move.

Third, and this one cuts against the instinct to treat courts as irrelevant in a weak-enforcement market. Post-communist evidence on the same question finds that belief in the effectiveness of courts has a significant positive effect on the level of trust shown in new relationships, and that workable courts encourage entrepreneurs to try new suppliers at all.⁴ The authors put the division plainly: relationships can sustain existing interactions, while courts help new interactions get started.⁴ So the patience inequality governs the relationships you already have. It does not, on its own, get you the first one.

Fourth, and the reason CHAIN gets run at all: what breaks these arrangements is rarely a change of heart. It is a shock that destroys future surplus. Political instability has been used precisely as such a shock, because it removes the future the relationship was priced against, and the relational contract breaks regardless of the history behind it.²

The reference class, the history lens

• Assumes: arrangements matched on structure fail at a knowable rate.
• Fits because: thousands of founders have given a counterparty this exact exposure.
• Breaks when: the rules changed, in which case the historical series is describing a different process.
• Evidence: grade A, with field identification from observed temptations to deviate rather than from stated intentions.
• Counteracts: the inside view, which builds the forecast from this deal’s features.
• May reinforce: a base rate imported from a market with different enforcement.

The four levers, cheapest first

Run the diagnosis and one of four variables will be binding. Pull that one. Pulling the others is effort with no mechanism behind it.

• If the horizon is binding, make the future visible and long before it becomes short. Renew the agreement in month nine of twelve, not in month twelve. Cost: a conversation. Reversible.
• If the grab is binding, shrink what any single period puts in their hands. Weekly settlement instead of monthly. Split the shipment. Cost: operational overhead. Reversible.
• If observability is binding, give them a second signal that separates bad luck from bad conduct. A shared tracker, a third-party inspection, a delivery photo with a timestamp. Cost: real but small. Reversible.
• If you have no relationship yet, the lever is not patience at all. Something external has to carry the first transaction: an escrow, a guarantee, a court you both believe works, or a third party with a record. The same evidence shows trust develops quickly once dealings begin, so the cost is concentrated at the start.⁴
• If enforcement is binding, replace expensive penalties with cheap ones. Withholding the next order costs you nothing to impose. Litigation costs you more than the breach. A penalty you would rather not enforce does not deter, because both sides can see you would rather not enforce it.

Note what is missing from that list. There is no lever called building the relationship, because it is not a variable. It is the name we give to the state where the inequality holds.

What to do before the next settlement date

Do now, and cap it. Write down the four numbers for your three largest counterparty exposures. Per-period value to them, the grab available, their visible horizon, and the enforcement you would actually use. This is reversible and it dominates across every scenario, so it goes first. Size it at an afternoon, not a project.

Hedge, and the premium is the whole loss. For the single largest grab on that list, halve the settlement period for one cycle. If the counterparty is sound you have lost a small amount of working-capital efficiency. That loss is the entire downside, which is what makes it a hedge rather than a bet.

Defer and trigger, with the size fixed now. Do not restructure the whole book of agreements today. Pre-commit the trigger instead: the first time a counterparty misses a settlement, the escalation schedule moves them back one step, automatically, without a meeting. Decide the size of that step now, while you are calm, rather than on the day it fires.

Each of those carries a cap. A cheap reversible action sized at a third of your runway is not cheap, and reversibility is not a defence against magnitude.

What history says happens next

Before running the base rate, run the break test, because a base rate drawn from a broken structure is worse than no base rate. Has a rule changed, has an actor entered or left, has a measurement become a target? If any of those fired, widen your bounds and stop projecting.

If none fired, the reference class is clear enough. Arrangements like this fail at the point where the future surplus is destroyed, not at the point where someone becomes dishonest. The failures cluster around shocks: a supply shock, an instability event, a competitor thickening the market and raising your counterparty’s outside option.

Subtract the counterfactual before you congratulate yourself on a long-standing relationship. A counterparty who has never defected in a period where defection was unattractive has told you nothing. What you learn from is a temptation they had and did not take, which is exactly the identification strategy the field literature uses to put a lower bound on relationship value.²

The part this ensemble is blind to

All three models take the counterparty as the object of analysis. None of them models you.

Your own horizon is visible to your suppliers. Your own signal to the market is noisy. If your runway is six months and your distributor can read that from your payment behaviour, then you are the player whose end is in sight, and every incentive described here runs in the other direction. The literature on this is honest about the limit too: reputation effects are temporary, because beliefs converge, and they persist only where uncertainty about type is continually replenished.¹ That applies to your reputation as much as to theirs.

There is also a limit on the whole apparatus. Sustainability is not selection. Showing that cooperation can hold does not tell you which arrangement emerges, and no amount of running this framework will produce that answer.¹

The one action that survives the ignorance: before your next settlement date, run the same four numbers with yourself as the counterparty and your largest supplier as the one deciding whether to extend you terms. Whatever comes out binding on that run is the thing to fix first, and it will not be the thing you were about to fix.

Who has to move

This only changes anything if the person who sets settlement terms is the person who reads it. In most companies that is finance, not the founder, and finance is optimising working capital rather than relationship sustainability. The cheapest first test is to take one counterparty, halve the settlement period for a single cycle, and see whether anything breaks. If nothing does, you have learned that the grab was larger than it needed to be, and you learned it for the price of one cycle.

Sources and notes

1. George J. Mailath and Larry Samuelson, Repeated Games and Reputations: Long-Run Relationships, Oxford University Press, 2006. The patience condition and minmax construction are developed in chapters 2 and 3; the Osborne and Rubinstein passage on perceived horizons is quoted at section 4.1; “Punishments happen” is section 7.2.1; the effect of increased monitoring precision is section 7.4; temporary reputations are section 15.5; the multiplicity problem is discussed at section 3.2, “Interpreting the Folk Theorem”.
2. Rocco Macchiavello and Ameet Morjaria, Relational Contracts: Recent Empirical Advancements and Open Questions, NBER Working Paper 30978, 2023. https://www.nber.org/system/files/working_papers/w30978/w30978.pdf. Covers the identification of relationship value from observed temptations to deviate, the Rwanda coffee finding on bilateral versus generalised trust, and the use of political instability as an exogenous shock to future surplus.
3. Rocco Macchiavello and Ameet Morjaria, The Value of Relationships: Evidence from a Supply Shock to Kenyan Rose Exports, University of Warwick Economic Research Paper 1032. https://wrap.warwick.ac.uk/id/eprint/59370/1/WRAP_twerp_1032_macchiavello.pdf. The abstract states that the value of the relationship increases with the age of the relationship, that sellers prioritise relationships during an exogenous negative supply shock, and that reliability at the time of the shock correlates with future survival and relationship value. Published in the American Economic Review.
4. Simon Johnson, John McMillan and Christopher Woodruff, Courts and Relational Contracts, Journal of Law, Economics and Organization 18(1), 2002. Author copy: https://chriswoodruff.qeh.ox.ac.uk/wp-content/uploads/2019/10/Courts_JLEO.pdf. The abstract states that belief in the effectiveness of courts has a significant positive effect on the level of trust shown in new relationships between firms and their customers, that well-functioning courts encourage entrepreneurs to try out new suppliers, and that while relationships can sustain existing interactions, workable courts help new interactions to develop. The paper also notes that the relevant survey measure is whether entrepreneurs believe the courts could be used if a dispute arose, precisely because the theory of repeated games models beliefs rather than realised disputes.

A note on what is not cited here. The folk theorem is often quoted as though it forecasts cooperation. It does not. It shows that with sufficiently patient players almost any individually rational payoff can be sustained, which is a statement about the size of the possible set and not a prediction about which point in it you will land on. This piece uses the inequality that generates the theorem and does not use the theorem.

Joshua Agonya Pi’Rwot, Founder.